Stress-Sensitive Controllability Inference

Computational model (Ligneul et al. 2020): Humans maintain parallel "actor" (controllable) and "spectator" (uncontrollable) models. The mPFC encodes specifically the prediction errors that disambiguate controllable from uncontrollable transitions. Controllability could be decoded from frontoparietal network patterns.

Key finding

Uncontrollable stressors bias controllability estimation mechanisms to promote reliance on the spectator model -- stress biases the inference system itself.

Positive feedback loop

Computational modeling (Karvelis & Diaconescu 2024) predicts this produces a positive feedback loop: stress -> biased controllability inference -> perceived uncontrollability -> more stress -> further bias. Whether this escalating loop operates in real time in humans has not been empirically demonstrated, but the directional bias is established.

Architectural implication

The controllability inference system is itself state-dependent. The stabilizer must interrupt this loop, not just deliver "controllable" experiences.

Source verification

Ligneul et al. 2020 (bioRxiv preprint, "Stress-sensitive inference of task controllability") -- verified.